Laplacian-Driven Softmax Probability for Noisy Pixel Detection in Image Reconstruction

The paper presents a modified tool for detection of noise in images when the pixel intensity of noise is within the pixel range. The detection is based on a combination of an edge-detection tool, i.e. the Laplacian filter, and the Softmax function. Although considered as an edge detection technique in the digital image processing, the Laplacian can be efficiently used in the determining noisy pixels. The decision–making threshold is then found using the Softmax function, which is based on the probability of the pixel intensity, which decides if a pixel should be declared as corrupted. The reconstruction is then performed using a gradient-based algorithm. The method is tested and verified through examples.